Jarek Potiuk

The root cause is here: https://github.com/apache/airflow/actions/runs/12743121090/job/35512486483?pr=44809 You need to run pre-commit with your change to regenerate .json file where we keep dependencies cross-providers.

> Ok my bad, how many times didn't I already forgot that one 😆 Thx @potiuk I actually never remember about it as well. I simply run `pre-commit install` and...

> @davidism Could you please weigh in? I think 8.3.0 makes sense. This one fixes #2952 - which is very annoying for CI - and reason why we have <...

BTW. We tested this one in Airflow and it looks good and indeed solves the #2952 issue. I can confirm that.

I do not know `pip` resiolution internals - but the rules explained make sense and might improve a number of cases indeed.

Almost. The solution in https://github.com/apache/airflow/pull/40272 is that we add >=2.1.2 for all our pandas dependencies - and this will at least skip 2.1.1 that causes numpy 2.0.0 from being considered...

For Python 3.8 we had to keep >= 1.5.3 though, because otherwise it would have conflict, but then Numpy 2 is Python 3.9+ only, so that is not an issue.

Not an easy one because nobody can guess far in advance what future will hold and whether certain future versions will be incompatible -especially that there are no universal standards...

> Look at BaseExecutor.log_task_event Nice.

Tried, seems it still does not work for airflow: 1) Checked out latest airflow (https://github.com/apache/airflow) 2) Installed latest hatch `pip install "hatch @ git+https://github.com/pypa/hatch"` 3) Double check: ``` [jarek:~/code/airflow] [apache-airflow]...